Research hardware options to train models more quickly

[acs]

After the meetup today at:

https://www.meetup.com/es-ES/MachineLearningSpain/events/260664781/

the nVidia GPUs option, using the Rapids framework which offers Pandas API and others implemented over GPUs, and the Edge TPUs from Google (Coral dev, the one used in the meetup demo) which can be used from Tensorflow, are interesting options to explore once we need howto improve the training of the ML models.

GTC (GPU Technical Conference) is a great place to explore AI topics: https://on-demand-gtc.gputechconf.com/gtcnew/on-demand-gtc.php (also deep learning eduation)

Deep Learning on the Edge

[acs]

Edge TPUs are mainly for the inference, not for the training:

https://coral.withgoogle.com/docs/edgetpu/faq/
"The Edge TPU is not capable of backward propagation"

For training probably it is better: https://www.nvidia.com/es-es/autonomous-machines/embedded-systems/jetson-nano/

[acs]

I have bought the nVidia Jetson Nano. It will arrive next 13th May. Not sure I could train models with it, but for sure, I will learn a lot, specially if I can mount it in the Jetbot.